idea topology and databases

> Well, I extremely tersely asked a speaker on Filemaker Pro if
> distributed hash tables came into his view and he said that was hidden
> in the machine he had bought and used . I didn't dare mention
> distributed garbage collection .
>
> I was afraid to ask if he had semantic parsers to natural language
> interface into his database or better built them himself . That would
> have been perceived as my showing off and being off topic .
>
> His business marketbase might not really approve of wondering how ideas
> are connected scientifically and maybe he has given himself over to not
> being inquisitive . In a different context, natural language questions
> aren't an embarrassment but a natural activity .
>
> I think I saw semantics or at least a parser with a grammar (rather than
> just taking dictation) was in voice recognition framework in os x. That
> was an exciting relief to crusty market to bottom line business
> dismissal (we're going to sell someone else's software to users rather
> than wonder how software is made or designed) .
>
> I have downloaded Fleury's pdf files to speedread. Since I'm not steeped
> in the subject, they might not inspire me, just let me know a bit what
> is going on and be able to listen better to what is going on , though I
> know this his heavy knowledge navigator stuff.

> On Sat, 2007-05-12 at 13:46 -0500, Paul Sheldon wrote:
> > Well, I extremely tersely asked a speaker on Filemaker Pro if
> > distributed hash tables came into his view and he said that was hidden
> > in the machine he had bought and used . I didn't dare mention
> > distributed garbage collection .
> >
> > I was afraid to ask if he had semantic parsers to natural language
> > interface into his database or better built them himself . That would
> > have been perceived as my showing off and being off topic .
> >
> > His business marketbase might not really approve of wondering how ideas
> > are connected scientifically and maybe he has given himself over to not
> > being inquisitive . In a different context, natural language questions
> > aren't an embarrassment but a natural activity .
> >
> > I think I saw semantics or at least a parser with a grammar (rather than
> > just taking dictation) was in voice recognition framework in os x. That
> > was an exciting relief to crusty market to bottom line business
> > dismissal (we're going to sell someone else's software to users rather
> > than wonder how software is made or designed) .
> >
> > I have downloaded Fleury's pdf files to speedread. Since I'm not steeped
> > in the subject, they might not inspire me, just let me know a bit what
> > is going on and be able to listen better to what is going on , though I
> > know this his heavy knowledge navigator stuff.
> I read both the papers, and they address specifically means and
> algorithm of searches, and while the word semantic is in the papers,
> their use of semantic is in reference to a centroid of the network, and
> uses an algorithm to calculate what they refer to as "semantic
> distance", which deals with the relevance of the data to the words in
> the search engine.  At least that is what my tiny mind interpreted from
> the papers.  If I am wrong, I hope someone will point out my error.
>
>     History, context, semantics and perception of meaning are vital to
> human discourse.  The processes of reducing these to some algorithm is
> difficult, and currently seems to be separate, at least in the minds of
> developers, which gives us lots of "almost there" interfaces (a great
> improvement from "not even close" in the 80's.)  Yet there is another
> aspect of the process, that of presentation, and in presentation, that
> of human capacity, sort of an ergonomics of human understanding or
> capacity.
>
>     These search engines deal with organizing the search space for speed
> of delivery, and in p2p (peer to peer, such as the music sharing sites,
> or the social sites, and even croquet spaces) the best and fastest
> search algorithm is one that runs the searches in parallel hitting some
> optimum number of immediate peers with some form of passing that search
> to another group, where the groups are typically referred to as clusters
> (bear with me, I know that many of you understand this, but this serves
> to help me set my thought and stage later stuff.)  Such algorithms are
> most efficient when the search depth remains short, and the number of
> active searches is high, thus latency is low and response fast.  But
> this means a physical representation would be something like a hallway
> with 10,000 doors, and the casual human browser would be overwhelmed by
> the options.  Ergonomics demands that the presentation be smaller, and
> that it be structured for best relevance to history, context, natural
> language semantics, and hopefully some degree of perception.
>
>     Perhaps a good visual might be an "Oracle" or "Teacher", where the
> participant could ask questions, the search algorithm could run and a
> count of retrieved responses could be used to determine available
> presentation level (say 6 or so, like the points of a good
> presentation).  If the search was too broad, the number of returned
> points would be great, and the "Oracle" would ask more questions and
> restate the search over the returned locations to reduce the noise
> before presenting options to the user.  There should be some limit to
> this interaction, and when the limit is reached, the available locations
> would be prioritized and the first N locations presented.  If these were
> insufficient, then another round of searches would be initiated against
> the remaining data set, or a new dataset could be generated.
>
>     This is just one thought, and I am not really happy with it.
>
>     I am by nature a browser of information, and a user both.  Thus
> there are times when I want a truely directed search, returning items
> that really match my criteria.  Amazon.com or Ebay is not an answer, but
> another place to browse for example.  If I want data on an electronic
> device, say ADC0808, then I type into a search engine "datasheet on
> ADC0809" should return the datasheets from the various manufacturers of
> that part (forgive me National).  On the other hand if I am looking to
> learn about new converter technologies, I might search for "data and
> algorithms for analog to digital converters", where Amazon.com may have
> a book that I should read, or Ebay may have some books for sale, and the
> papers of the IEEE or ACM or the Robotics Institute (if it exists) might
> have information relevant to the issue.  However a good search assistant
> would ask "do you want to include commercial sites for books?" and I
> could answer yes or no to aid in providing the desired information.  If
> the search returned far too many sites, it might ask "do you want to see
> sales sites for converters" and I could answer no.
> Thus it could help filter the data.  Yes, I could put all the filter
> information into the search query, but when I have to do that there is
> data I might overlook or relationships that I might not realize without
> reading the data from the site.  I would like the systems to help
> restore balance.  I do not want to see Brittany Spears in my search
> results, thank you.  This is not something that will be
> implemented immediately, but will take evolution on a computer scale and
> via open development to bring about.
>
> Then once the information has parsed down to a reasonable number of
> potential targets, how would those be presented?  And importantly how
> fast will this process be?
>
> Am I making any sense at all?
>
> Regards,
> Les H
>
>